Computer Vision News - March 2023

44 WASPSYN Challenge at ISBI morphology in different brain regions . “ We want to apply our model to the whole brain or even another brain, so we require the trained model to be generalizable to many brain regions and datasets, ” he points out. “ Then, these methods can be applied in real-world connectomics. The CREMI methods don’t have this guarantee. ” Solving this problem proves challenging due to the extensive variability of the neuron morphology in brain regions . Furthermore, there is a lack of consistency even when examining different brain samples with the same microscope. Illumination and sample preparation can to another neuron from the pre-synapse to post-synapse, ” he explains. “ It has a direction from neuron A to neuron B. In this challenge, we want to detect the synapses automatically and accurately. ” WASPSYN is distinct froman existing circuit reconstruction challenge called CREMI . The primary difference lies in the type of electron microscopy used for imaging. In CREMI, physical sectioning is employed, resulting in a section thickness of 40 nm, compared to 8 nm here, and an XY lateral resolution of 4 nm. Also, the training and test volumes are very close. The image texture is similar, which Jingpeng says does not reflect the diverse neuron Baseline method using 3D U-Net. (a) Training of T-bar detection network. The patches are randomly sampled (b) Training of post-synapse detection network. The T-bar is in the center of each patch and a fixed patch with a cen the illustration is 2D while both the patches and network are 3D and of (a) Task 1 (b) Task 2 Presynapse Image Image seline method using 3D U-Net. (a) Training of T-bar detection network. The patch any T-bar. (b) Training of post-synapse detection n twork. The T-bar is in the cente a channel of input patch. Note that the illustration is 2D while both the patches and